Théoriede la décisionet théoriedes jeux Master MODO …moretti/teaching/core_and_Shapley.pdf ·...
Transcript of Théoriede la décisionet théoriedes jeux Master MODO …moretti/teaching/core_and_Shapley.pdf ·...
1
An introduction to
cooperative gamesStefano Moretti
UMR 7243 CNRS
Laboratoire d'Analyse et Modélisation de Systèmes pour
l'Aide à la décision (LAMSADE)
Université Paris-Dauphine
email: [email protected] room: P422
Théorie de la décision et théorie des jeux
Master MODO 2016-2017
�Dominant strategies
�Nash eq. (NE)
�Subgame perfect NE
�NE & refinements
…
�Core
�Shapley
value
�Nucleolus
�τ-value
�PMAS
….
�Nash sol.
�Kalai-
Smorodinsky
….
�CORE
�NTU-value
�Compromise
value
…
No binding agreements
No side payments
Q: Optimal behaviour in conflict
situations
binding agreements
side payments are possible (sometimes)
Q: Reasonable (cost, reward)-sharing
2
Cooperative games: a simple exampleAlone, player 1 (singer) and 2 (pianist) can
earn 100€ 200€ respect.
Together (duo) 700€
How to divide the (extra) earnings?
I(v)700
400
600
200
100 300 500 700
x2
x1x1 +x2=700
Imputation set: I(v)={x∈IR2|x1≥100, x2≥200, x1+x2 =700}
Example: ice-cream game
(N,v) such that N={1,2,3}
v(1)=v(3)=0
v(2)=3
v(1,2)=3
v(1,3)=1
v(2,3)=4
v(1,2,3)=5
Interpretation: kids want to buy ice-cream
1 and 3 cooperate
1 and 3 not enough money � no ice-cream
1 Kg
2 alone 3 Kg
1 and 2 cooperate 3 Kg
2 and 3 cooperate 4 Kg
1,2 and 3 cooperate 5 Kg
Q.1: will do they join their money
In order to buy more ice-cream?
Q.2: if yes, how do they share the
Ice-cream?
3
COOPERATIVE GAME THEORY
Games in coalitional form
TU-game: (N,v)
N={1, 2, …, n} set of players
S⊂N coalition
2N set of coalitions
v: 2N�IR characteristic function
v(S) is the value (worth) of coalition S
DEF. (N,v) with v(∅)=0 is a Transferable Utility (TU)-game with
player set N.
NB: if N is fixed, (N,v)↔v
NB: if n=|N|, it is also called n-person TU-game…
Other names: game in coalitional form, coalitional game,
cooperative game with side payments...
Q.1: which coalitions form?
Q.2: If the grand coalition N forms, how to divide v(N)?
(how to allocate the utility or the cost of the grand coalition?)
Many answers! (solution concepts)
One-point concepts: - Shapley value (Shapley 1953)
- nucleolus (Schmeidler 1969)
-τ-value (Tijs, 1981)
…
Subset concepts: - Core (Gillies, 1954)
- stable sets (von Neumann, Morgenstern, ’44)
- kernel (Davis, Maschler)
- bargaining set (Aumann, Maschler)
…..
DEF. (N,v) is a superadditive game iff
v(S∪T)≥v(S)+v(T) for all S,T with S∩T=∅
4
Claim: (N,v) is superadditive
We show that v(S∪T)≥v(S)+v(T) for all S,T∈2N\{∅} with S∩T=∅
3=v(1,2)≥v(1)+v(2)=0+3
1=v(1,3)≥v(1)+v(3)=0+0
4=v(2,3)≥v(2)+v(3)=3+0
5=v(1,2,3) ≥v(1)+v(2,3)=0+4
5=v(1,2,3) ≥v(2)+v(1,3)=3+1
5=v(1,2,3) ≥v(3)+v(1,2)=0+3
Example: ice-cream game
(N,v) such that N={1,2,3}
v(1)=v(3)=0
v(2)=3
v(1,2)=3
v(1,3)=1
v(2,3)=4
v(1,2,3)=5
The imputation set
DEF. Let (N,v) be a n-persons TU-game.
A vector x=(x1, x2, …, xn)∈IRN is called an imputation iff
(1) x is individual rational i.e.
xi ≥ v(i) for all i∈N
(2) x is efficient
Σi∈N xi = v(N)
[interpretation xi: payoff to player i]
I(v)={x∈IRN | Σi∈N xi = v(N), xi ≥ v(i) for all i∈N}
Set of imputations
5
Example
(N,v) such that
N={1,2,3},
v(1)=v(3)=0,
v(2)=3,
v(1,2,3)=5.
x3
x2
X1
(5,0,0)
(0,5,0)
(0,0,5)
x1+x2
+x3=5
(2,3,0)
(0,3,2)
I(v)
I(v)={x∈∈∈∈IR3 | x1,x3≥≥≥≥0, x2≥≥≥≥3, x1+x2+x3=5}
5 Kg
3 Kg2 Kg
, , )(
player1 player2 player3
The core of a game
DEF. Let (N,v) be a TU-game. The core C(v) of (N,v) is the
set
C(v)={x∈I(v) | Σi∈S xi ≥ v(S) for all S∈2N\{∅}}
stability conditions
no coalition S has the incentive to split off if
x is proposed
Note: x ∈ C(v) iff
(1) Σi∈N xi = v(N) efficiency
(2) Σi∈S xi ≥ v(S) for all S∈2N\{∅} stability
Bad news: C(v) can be empty
Good news: many interesting classes of games have a non-
empty core.
6
Example
(N,v) such that
N={1,2,3},
v(1)=v(3)=0,
v(2)=3,
v(1,2)=3,
v(1,3)=1
v(2,3)=4
v(1,2,3)=5.
Core elements satisfy the
following conditions:
x1,x3≥0, x2≥3, x1+x2+x3=5
x1+x2≥3, x1+x3≥1, x2+x3≥4
We have that
5-x3≥3⇔x3≤2
5-x2≥1⇔x2≤4
5-x1≥4⇔x1≤1
C(v)={x∈IR3 | 1≥x1≥0,2≥x3≥0, 4≥x2≥3, x1+x2+x3=5}
Example
(N,v) such that
N={1,2,3},
v(1)=v(3)=0,
v(2)=3,
v(1,2)=3, v(1,3)=1
v(2,3)=4
v(1,2,3)=5.
x3
x2
X1
(5,0,0)
(0,5,0)
(0,0,5)
x1+x2
+x3=5
(2,3,0)
(0,3,2)
C(v)(0,4,1)
(1,3,1)
(1,4,0)
C(v)={x∈IR3 | 1≥x1≥0,2≥x3≥0, 4≥x2≥3, x1+x2+x3=5}
7
Example (Game of pirates) Three pirates 1,2, and 3. On the other
side of the river there is a treasure (10€). At least two pirates are
needed to wade the river…
(N,v), N={1,2,3}, v(1)=v(2)=v(3)=0,
v(1,2)=v(1,3)=v(2,3)=v(1,2,3)=10
Suppose (x1, x2, x3)∈C(v). Then
efficiency x1+ x2+ x3=10
x1+ x2 ≥10
stability x1+x3 ≥10
x2+ x3 ≥10
20=2(x1+ x2+ x3) ≥30 Impossible. So C(v)=∅.
Note that (N,v) is superadditive.
Example
(Glove game with L={1,2}, R={3})
v(1,3)=v(2,3)=v(1,2,3)=1, v(S)=0 otherwise
Suppose (x1, x2, x3)∈C(v). Then
x1+ x2+ x3=1 x2=0
x1+x3 ≥1 x1+x3 =1
x2≥0
x2+ x3 ≥1 x1=0 and x3=1
So C(v)={(0,0,1)}.
(1,0,0)
(0,0,1)
(0,1,0)
I(v)
8
How to share v(N)…
�The Core of a game can be used to exclude those
allocations which are not stable.
�But the core of a game can be a bit “extreme” (see
for instance the glove game)
�Sometimes the core is empty (pirates)
�And if it is not empty, there can be many
allocations in the core (which is the best?)
An axiomatic approach (Shapley (1953)
�Similar to the approach of Nash in bargaining:
which properties an allocation method should
satisfy in order to divide v(N) in a reasonable way?
�Given a subset C of GN (class of all TU-games with
N as the set of players) a (point map) solution on C
is a map Φ:C→IRN.
�For a solution Φ we shall be interested in various
properties…
9
Symmetry
PROPERTY 1(SYM) Let v∈GN be a TU-game.
Let i, j∈Ν. If v(S∪{i}) = v(S∪{j}) for all S∈2N\{i,j},
then Φi(v) = Φj (v).
EXAMPLE
We have a TU-game ({1,2,3},v) s.t. v(1) = v(2) = v(3) = 0,
v(1, 2) = v(1, 3) = 4, v(2, 3) = 6, v(1, 2, 3) = 20.
Players 2 and 3 are symmetric. In fact:
v(∅∪{2})= v(∅∪{3})=0 and v({1}∪{2})=v({1}∪{3})=4
If Φ satisfies SYM, then Φ2(v) = Φ3(v)
Efficiency
PROPERTY 2 (EFF) Let v∈GN be a TU-game.
∑ i∈NΦi(v) = v(N), i.e., Φ(v) is a pre-imputation.
Null Player Property
DEF. Given a game v∈GN, a player i∈N s.t.
v(S∪i) = v(S) for all S∈2N will be said to be a null player.
PROPERTY 3 (NPP) Let v∈GN be a TU-game. If i∈N is a
null player, then Φi(v) =0.
EXAMPLE We have a TU-game ({1,2,3},v) such that v(1)
=0, v(2) = v(3) = 2, v(1, 2) = v(1, 3) = 2, v(2, 3) = 6, v(1, 2,
3) = 6. Player 1 is null. ThenΦ1(v) = 0
10
EXAMPLE We have a TU-game ({1,2,3},v) such that
v(1) =0, v(2) = v(3) = 2, v(1, 2) = v(1, 3) = 2, v(2, 3)
= 6, v(1, 2, 3) = 6. On this particular example, if Φ
satisfies NPP, SYM and EFF we have that
Φ1(v) = 0 by NPP
Φ2(v)= Φ3(v) by SYM
Φ1(v)+Φ2(v)+Φ3(v)=6 by EFF
So Φ=(0,3,3)
But our goal is to characterize Φ on GN. One more
property is needed.
Additivity
PROPERTY 4 (ADD) Given v,w ∈GN,
Φ(v)+Φ(w)=Φ(v +w).
.EXAMPLE Two TU-games v and w on N={1,2,3}
v(1) =3
v(2) =4
v(3) = 1
v(1, 2) =8
v(1, 3) = 4
v(2, 3) = 6
v(1, 2, 3) = 10
w(1) =1
w(2) =0
w(3) = 1
w(1, 2) =2
w(1, 3) = 2
w(2, 3) = 3
w(1, 2, 3) = 4
+
v+w(1) =4
v+w(2) =4
v+w(3) = 2
v+w(1, 2) =10
v+w(1, 3) = 6
v+w(2, 3) = 9
v+w(1, 2, 3) = 14
=
Φ ΦΦ
11
Theorem 1 (Shapley 1953)
There is a unique map φ defined on GN that satisfies EFF,
SYM, NPP, ADD. Moreover, for any i∈N we have that
)(!
1)( vm
nv ii ∑ Π∈
=σ
σφ
Here Π is the set of all permutations σ:N→N of N, while mσi(v) is
the marginal contribution of player i according to the permutation σ,
which is defined as:
v({σ(1), σ(2), . . . , σ (j)})− v({σ(1), σ(2), . . . , σ (j −1)}),
where j is the unique element of N s.t. i = σ(j).
Example
(N,v) such that
N={1,2,3},
v(1)=v(3)=0,
v(2)=3,
v(1,2)=3,
v(1,3)=1,
v(2,3)=4,
v(1,2,3)=5.
Permutation 1 2 3
1,2,3 0 3 2
1,3,2 0 4 1
2,1,3 0 3 2
2,3,1 1 3 1
3,2,1 1 4 0
3,1,2 1 4 0
Sum 3 21 6
φ(v) 3/6 21/6 6/6
Probabilistic interpretation: (the “room parable”)
�Players gather one by one in a room to create the “grand coalition”, and each
one who enters gets his marginal contribution.
�Assuming that all the different orders in which they enter are equiprobable,
the Shapley value gives to each player her/his expected payoff.
12
Example
(N,v) such that
N={1,2,3},
v(1)=v(3)=0,
v(2)=3,
v(1,2)=3, v(1,3)=1
v(2,3)=4
v(1,2,3)=5.
x3
x2
X1
(5,0,0)
(0,5,0)
(0,0,5)
x1+x2
+x3=5
(2,3,0)
(0,3,2)
C(v)
Marginal vectors
123�(0,3,2)
132�(0,4,1)
213�(0,3,2)
231�(1,3,1)
321�(1,4,0)
312�(1,4,0)
(0,4,1)
(1,3,1)
(1,4,0)
φ φ φ φ(v)=(0.5, 3.5,1)
Unanimity games (1)
�DEF Let T∈2N\{∅}. The unanimity game on T is defined as the TU-game (N,uT) such that
1 is T⊆S
uT(S)=
0 otherwise
�Note that the class GN of all n-person TU-games is a vector space (obvious what we mean for v+w and αv for v,w∈GN and α∈IR).
� the dimension of the vector space GN is 2n-1
�{uT|T∈2N\{∅}} is an interesting basis for the vector space GN.
13
Unanimity games (2)
�Every coalitional game (N, v) can be written as a
linear combination of unanimity games in a unique
way, i.e., v =∑S∈2NλS(v)uS .
�The coefficients λS(v), for each S∈2N, are called
unanimity coefficients of the game (N, v) and are
given by the formula: λS(v) = ∑T∈2S (−1)s−t v(T ).
Sketch of the Proof of Theorem1
�Shapley value satisfies the four properties (easy).
�Properties EFF, SYM, NPP determine φ on the class
of all games αv, with v a unanimity game and
α∈IR.
�Let S∈2N. The Shapley value of the unanimity game
(N,uS) is given by
α/|S| if i∈S
φi(αuS)=
0 otherwise
�Since the class of unanimity games is a basis for the
vector space, ADD allows to extend φ in a unique
way to GN.
14
�Let mσ’i(v)=v({σ’(1),σ’(2),…,σ’(j)})− v({σ’(1),σ’(2),…, σ’(j −1)}),
where j is the unique element of N s.t. i = σ’(j).
�Let S={σ’(1), σ’(2), . . . , σ’(j)}.
�Q: How many other orderings σ∈Π do we have in which
{σ(1), σ(2), . . . , σ (j)}=S and i = σ’(j)?
�A: they are precisely (|S|-1)!×(|N|-|S|)!
�Where (|S|-1)! Is the number of orderings of S\{i} and (|N|-
|S|)! Is the number of orderings of N\S
�We can rewrite the formula of the Shapley value as the
following:
( )∑ ∈∈−
−−=
SiSi N iSvSvn
snsv
:2}){\()(
!
)!()!1()(φ
An alternative formulation
Convex games (1)
DEF. An n-persons TU-game (N,v) is convex iff
v(S)+v(T)≤v(S ∪T)+v(S∩T) for each S,T∈2N.
This condition is also known as supermodularity. It can be
rewritten as
v(T)-v(S∩T)≤v(S ∪T)-v(S) for each S,T∈2N
For each S,T∈2N, let C=(S∪T)\S. Then we have:
v(C∪(S∩T))-v(S∩T)≤v(C ∪S)-v(S)
Interpretation: the marginal contribution of a coalition C to a
disjoint coalition S does not increase if S becomes smaller
15
Convex games (2)�It is easy to show that supermodularity is equivalent to
v(S ∪{i})-v(S)≤v(T∪{i})-v(T)
for all i∈N and all S,T∈2N such that S⊆T ⊆ N\{i}
�interpretation: player's marginal contribution to a large
coalition is not smaller than her/his marginal contribution to a
smaller coalition (which is stronger than superadditivity)
�Clearly all convex games are superadditive (S∩T=∅…)
�A superadditive game can be not convex (try to find one)
�An important property of convex games is that they are
(totally) balanced, and it is “easy” to determine the core
(coincides with the Weber set, i.e. the convex hull of all
marginal vectors…)
Example
(N,v) such that
N={1,2,3},
v(1)=v(3)=0,
v(2)=3,
v(1,2)=3, v(1,3)=1
v(2,3)=4
v(1,2,3)=5.
Check it is convex
x3
x2
X1
(5,0,0)
(0,5,0)
(0,0,5)
x1+x2
+x3=5
(2,3,0)
(0,3,2)
C(v)
Marginal vectors
123�(0,3,2)
132�(0,4,1)
213�(0,3,2)
231�(1,3,1)
321�(1,4,0)
312�(1,4,0)
(0,4,1)
(1,3,1)
(1,4,0)
C(v)={x∈IR3 | 1≥x1≥0,2≥x3≥0, 4≥x2≥3, x1+x2+x3=5}
16
Section 2. Connection situations
31
http://www.vrtuosi.com
� A connection situation takes place in the presence of a group of agents N={1,2, …,n}, each of which needs to be connected directly or via other agents to a source.
� If connections among agents are costly, then each agent will evaluate the opportunity of cooperating with other agents in order to reduce costs.
� If a group of agents decides to cooperate, a configuration of links which minimizes the total cost of connection is provided by a minimum cost spanning tree (mcst).
� The problem of finding a mcst may be easily solved thanks to different algorithms proposed in literature (Boruvka (1926), Kruskal(1956), Prim (1957), Dijkstra (1959))
Minimum Cost Spanning Tree Situations
Consider a complete weighted graph
1
2
3
– whose vertices represent agents
source
– vertex 0 is the source
0– edges represent connections between
agents or between an agent and the source
40
30
10
50
20
– numbers close to edges are connection costs
80
17
Minimum cost spanning tree (mcst) problem
Optimization problem:
How to connect each node to the source 0 in
such a way that the cost of construction of a
spanning network (which connects every node
directly or indirectly to the source 0) is
minimum?
ExampleN={1,2,3},EN’={{1,0},{2,0},{2,1},{3,0},{3,1},{3,2}}
cost function shown on graphs
2
1
0
18
2424
26
10 20
3
Kruskal algorithm
2
1
0
18
2424
26
10
3
Prim algorithm
20
18
2
1
0
18
2424
26
10 20
3
c(1)=24
c(3)=26
c(2)=24
c(1,3)=34
c(2,3)=44
c(1,2)=42
c(1,2,3)=52
Example: mcst cost game ({1,2,3},c) defined on the
following connection situation:
2
1
0
18
2424
26
10 20
3
Example: The cost game ({1,2,3},c) is defined on the
following connection situation:
The game ({1,2,3}, c) is said mcst game (Bird (1976))
c(1)=24
c(2)=24
c(3)=26
c(1,3)=34
c(1,2)=42
c(2,3)=44
c(1,2,3)=52
19
2
1
0
18
2424
26
10 20
3
• The predecessor of 1 is 0: the Bird
allocation gives to player 1 the cost of {0,1}.
•The predecessor of 2 is 1: the Bird allocation gives to player 2 the cost of {1,2};
• The predecessor of 3 is 1: the Bird allocation gives to player 3 the cost of {1,3}.
w(Γ)=52
Bird allocation w.r.t. to Γ, (x1, x2, x3)=(24, 18 ,10) is in the core of ({1,2,3},c).
How to divide the total cost? (Bird 1976)
The Bird allocation w.r.t. this mcst is
(x1, x2, x3)=(18, 24 ,10)
The Bird allocation w.r.t .this
mcst is
(x1, x2, x3)=(24, 18 ,10)
2
1
0
18
2424
26
10 20
3
2
1
0
18
2424
26
10
3
20
Both allocations belong to the core of the mcst game (and
also their convex combination).
20
2
1
0
18
2424
26
10 20
3
(24,24,4)
(2,24,26)(24,2,26)
I(N,c)
(x1,x2,x3) s.t.
x1+x2+x3=52
x1≤24
x2 ≤24
x3 ≤26(18,24,10)(24,18,10)
(8,18,26)
Core(N,c)
(8,24,20)
x1+x2≤42
x2+x3≤44
x1+x2≤34
Bird 1 Bird 2
Bird allocation rule
� It always provides an allocation (given a connection situation).
� In general, not a unique allocation (each mcst determines a corresponding Bird allocation…).
� Bird allocations are in the core of mcstgames (but are extreme points)
21
What happens when the structure of the
network changes?
� Imagine to use a certain rule to allocate costs.
� The cost of edges may increase: if the cost of an edge
increases, nobody should be better off, according to such
a rule (cost monotonicity);
� One or more players may leave the connection situation:
nobody of the remaining players should be better off
(population monotonicity).
Cost monotonicity: Bird allocation behaviour
2
1
0
3
45
8
3 4
3
2
1
0
3
65
8
3 4
3
Bird allocation: (4, 3 ,3) Bird allocation: (3, 5 ,3)
Bird rule does not satisfy cost monotonicity.
22
Population monotonicity: Bird allocation
behaviour
Bird allocation: (5, 5 ,3) Bird allocation: (3, * ,6)
2
1
0
5
75
6
3 7
3
1
0
7
6
3
3
Bird rule does not satisfy population monotonicity
Construct & Charge rules
Are based on the following general cost allocation
protocol:
� As soon as a link is constructed in the Kruskal algorithm
procedure:
1) it must be totally charged among agents which are not yet
connected with the source (connection property)
2) Only agents that are on some path containing the new edge
may be charged (involvement property)
� when the construction of a mcst is completed, each agent
has been charged for a total amount of fractions equal to 1
(total aggregation property).
23
2
1
0
18
2424
26
10 20
3
2
1
0
3
,0 (1,1,1)tb
σ =
(0,0,0)
There are no
edge costs to
share.
2
1
0
3
,1 1 1( ,1, )2 2
tbσ =
10
(5,0,5)
1 and 3 share
cost 10
equally.
P-value: Feltkamp (1994), Branzei et al. (2004), Moretti (2008)
2
1
0
3
,2 1 1 1( , , )3 3 3
tb
σ =
(3,12,3)
18
2 is connected to 1 and
3 who were already
connected: 2 pays 2/3
of 18 whereas the
remaining is shared
equally between 1 and
3.
2
1
0
3
,3 1 1 1( , , )3 3 3
tb
σ =
(0,0,0)20
Oops… there is
a cycle: nobody
want it.
2
1
0
3,4 (0,0,0)t
bσ =
24 (8,8,8)
Players are
connected to
0: share the
total cost of
the last edge
(=24) equally
24
P-value
2
1
0
18
2424
26
10 20
3
Make the sum of all edge-by-edge allocations:
(0, 0, 0) +
(5, 0, 5) +
(3,12,3) +
(0, 0, 0) +(8, 8, 8) =
P-value = (16,20,16)
Algorithm to calculate the P-value
� At any step of the Kruskal algorithm where a component is connected to some agents, charge the cost of that edge among these agents in the following way:
� Proportionally to the cardinality_current_step-1 if an agent is connected to a component which is connected to the source,
� Otherwise, charge it proportionally to the difference: cardinality_previous_step-1 - cardinality_current_step-1
IDEA: charge the cost of an edge constructed during the
Kruskal algorithm only between agents involved, keeping into
account the cardinality of the connected components at that step
and at the previous step of the algorithm:
25
P-value
�Always provides a unique allocation (given a mcst
situation).
�It is in the core of the corresponding mcst game.
�Satisfies cost monotonicity.
�Satisfies population monotonicity.
�on a subclass of connection problems it coincides with the Shapley value of mcst games
�…
Weighted Majority Games
Suppose that four parties receive these vote shares:
Party A, 27%; Party B, 25%; Party C, 24%; Party 24%.
Seats are apportioned in a 100-seat parliament according some apportionment formula. In this case, the apportionment of seats is straight-forward:
Party A: 27 seats Party C: 24 seats
Party B: 25 seats Party D: 24 seats
Suppose a simple majority is required (at least 51 seats) to be winning
26
The Shapley-Shubik IndexSuppose coalition formation starts at the top of each ordering, moving
downward to form coalitions of increasing size.
At some point a winning coalition formed.
The “grand coalition” {A,B,C,D} is certainly winning.
For each ordering, identify the pivotal player who, when added to the players already in the coalition, converts a losing coalition into a winning coalition.
Given the seat shares of parties A, B, C, and D before the election, the pivotal player in each ordering is identified by the arrow (<=).
The Shapley-Shubik Index (cont.)
Voter i’s Shapley-Shubik power index value SS(i) is simply:
Number of orderings in which the voter i is pivotal Total number of orderings
Clearly such power index values of all voters add up to 1.
Counting up, we see that A is pivotal in 12 orderings and each of B, C, and D is pivotal in 4 orderings. Thus:
Voter SS PowerA 1/2 = .500B 1/6 = .167C 1/6 = .167 D 1/6 = .167
So according to the Shapley-Shubik index, Party A has 3 times the voting power of each other party.
Lloyd Shapley and Martin Shubik, American Political Science Review, 1955.
27
UN Security Council
• 15 member states:
– 5 Permanent members: China, France, Russian
Federation, United Kingdom, USA
– 10 temporary seats (held for two-year terms )
(http://www.un.org/)
UN Security Council decisions
• Decision Rule: substantive resolutions need the
positive vote of at least nine Nations but…
…it is sufficient the negative vote of one among
the permanent members to reject the decision.
• How much decision power each Nation inside the
ONU council to force a substantive decision?
• Game Theory gives an answer using the Shapley-
Shubik power index:
28
UN Security Council as a weighted majority game
�Let N=P∪T, where P={1,2,3,4,5} is the set of
Permanent members and
T={6,7,8,9,10,11,12,13,14,15} is the set of
temporary seats
�A simple game (N,v) s.t. v(S)=1 if |S|≥9 and P⊂S.
�(N,v) is a weighted majority game, where
�wi=7 for each i∈P
�wi=1 for each i∈T
�v(S)=1 ⇔ ∑i∈S wi≥ 39
Power≅ 19.6%
Power≅ 0.2%
temporary seat 2015-2016
Shapley-Shubik index
29
United States presidential elections
� are indirect elections in which voters will select presidential electors who in turn will elect a new president and vice president through the electoral college.
� electors directly elect the President (and Vice President).
�Once chosen, the electors can vote for anyone, but – with rare exceptions– they vote for their designated candidates
� Each state is allocated a number of Electoral College electors equal to the number of its Senators and Representatives in the U.S. Congress (i.e. by population).
�Additionally, Washington, D.C. is given a number of electors equal to the number held by the smallest state.
United States presidential election (3)
�A game (N,v), where |N|=51 (the number of states plus
Washington D.C.), each one with a weight given by the
number of electors
�v(S)=1 ⇔ ∑i∈S wi > 270
�In 1977, weights are between 3 (smallest states
and Washington D.C.) and 45 (California)
0.032020.007234Montaa
0.137360.0314717Florida
0.024020.0054123Washington DC
0.386940.0883145California
BanzhafShapleyElectorsState
30
United States presidential elections
� are indirect elections in which voters will select presidential electors who in turn will elect a new president and vice president through the electoral college.
� In this indirect election the power of each voter in California and Montana is the following
≅ 7.8476 x 10-9
≅ 3.4516 x 10-9
Owen, G. (1995) Game Theory. 3rd edn. Academic Press. Calculations of indices for the situation in 1970
Power indices: a general formulation
Let (N,v) be a simple game (assume v is monotone: for each
S,T ∈2N. S⊆T⇒ v(S) ≤v(T))
Let pi(S), for each S∈2N\{∅}, i∉S, be the probability of
coalition S∪{i} to form (of course ∑S⊆N:i∉S pi(S)=1)
A power index ψi(v) is defined as the probability of player i
to be pivotal in v according to p:
ψip(v)=∑S⊆N:i∉S pi(S) [v(S∪{i})-v(S)]